Do Vehicle Emissions Testing Program Improve Air Quality?
Transcript of Do Vehicle Emissions Testing Program Improve Air Quality?
Do Vehicle Emissions Testing ProgramImprove Air Quality?
Matthew E. Kahn, Columbia University
Discussion Paper No. 687
DO VEHICLE EMISSIONS TESTING PROGRAM IMPROVE AIR QUALITY?
Matthew E. Kahn
Columbia University
Department of Economics
February 1994
I thank Dora Costa and two anonymous referees for helpful comments. All mistakes are mine.
Abstract
This paper estimates the impact of the Illinois vehicle emissions testing program on ambient air quality.I find that this programs has reduced ozone but has had a smaller than expected impact on carbonmonoxide levels. Unlike previous environmental regulation evaluation studies, I control for the fact thatregulation is not randomly assigned across geographical areas.
I. Introduction
Federal regulation aimed at improving air quality has increased over the last twenty years.
Starting with the Clean Air Act of 1970, counties whose air quality did not meet federal standards have
been subject to increasingly stringent regulation on stationary and mobile sources. The Clean Air Act
Amendments of 1977 required states to implement emissions testing programs in counties not in
attainment with the ozone and carbon monoxide standards by 1982. The most recent regulation, the Clean
Air Act Amendments of 1990, call for even more stringent emissions testing as well as other measures.1
Emissions testing programs are the major regulation for reducing used cars' emissions.
Little research has studied the impact of federal regulation on air quality. Most evaluations have
been done by the regulatory agency. In one of the few studies done by economists, MacAvoy (1987)
found that regulation lowers air quality. But, the Environmental Protection Agency (EPA) claims that
emissions testing programs are responsible for improvements of 25% in Chicago's carbon monoxide
emissions levels.
Vehicle emissions testing program provide almost a natural experiment to evaluate regulation's
impact on air quality. Within a given state, regulation varies across counties. A county either has or does
not have a program. Each car that is tested within a county has the same test performed. However,
regulation is not randomly imposed. Since emissions testing is more likely to occur in heavily polluted
areas, failure to account for its endogeneity understates regulation's impact. MacAvoy's (1987) findings
could be explained by this.
This paper adds to the environmental regulation evaluation literature by explicitly taking into
account the regulation's endogeneity. I estimate the effect of the Illinois vehicle emissions testing
JSuch steps include; improved gasoline vapor recovery controls at service stations in 1993 and on carsin 1998. More stringent tail pipe emissions standards in 1994 car models and a cleaner gasoline requiredin most Northeast urban areas starting in 1995.
programs on air quality. No study has attempted to estimate how actual county air quality co-moves with
regulation.
The empirical strategy is to create a panel data set relating a county's yearly air quality measure
to weather and regulation variables. Given air quality data for counties that never implement an emissions
testing program, and pre- and post-intervention data for counties that do implement programs, I estimate
how the presence of the program affected county air quality. I show that because of program
implementation delays, a fixed effect estimator yields unbiased estimates of the program's impact.
This paper is organized as follows. Section Two presents some information about the Illinois
vehicle emissions testing program. Section Three presents a model relating vehicle regulation to county
air quality. Section Four presents my estimation strategy. Section Five presents my data sources. Section
Six presents the results. Section Seven concludes.
II. The Illinois Vehicle Emissions Testing Program
The typical emissions test procedure is designed to adjust and repair or replace common tune-up
components of the engine. The emissions test is discussed in more detail in the General Accounting Office
report (GAO 1992). In the state of Illinois five counties started emissions testing in 1986. The regulation
requires registered vehicle owners to bring their cars to be tested every two years. Vehicles that fail the
test are expected to be repaired and re-tested. The emissions test has different passing criteria by model
year. Older vehicles face more lax emissions standards. Table Two presents evidence on the passing
criteria by model year. Note that pass rates are roughly 80% per year. 1981 model year cars had the
lowest pass rates. Table Two also presents evidence that hydrocarbon and carbon monoxide emissions
have been declining with respect to model year.
State programs can differ on the frequency of testing, passing criteria, on the maximum
expenditure for receiving an emissions waiver ($500 is the upper bound on repair expenditure), and on
whether testing can occur at repair shops or not. In Illinois, emissions testing occurs at centralized
locations. Unlike in California, vehicle repair shops cannot test a vehicle's emissions. The EPA has
argued that these centralized testing programs are less likely to suffer from bribing problems than
decentralized repair shop testing (see the California I/M Review 1992).
III. An Aggregation Model of Vehicle Emissions' Impact on Air Quality
Abstracting from cross-county pollution spillovers, county air quality is a function of aggregate
economic activity within a county. Environmental regulation affects air quality by changing the level of
pollution per unit of economic activity and the level of economic activity within a given county. In 1985,
there was no emissions testing in Illinois. Suppose that in 1985 every car's emissions can take on one of
M values, where M is finite. Define the probability distribution over these M values as f(E);
Define N as the number of cars of type i. Assume that each driver drives 1 mile a year and that driving
is the only source of air pollution. Aggregate emissions in each county equals;
total emissions - J ^ N{ * J{Et) (2)
This aggregate emissions is an input, along with weather conditions, in determining actual county air
quality.
A " G(5Z!Ni */&},&) (3)
"A" represents county air quality and theta represents weather conditions. The environmental regulator's
problem is to design incentives such that total emissions map through equation (3) to generate a level of
air quality that meets the Clean Air Act standards.
My focus is to study how the presence of a vehicle emissions testing program affects total
emissions, equation (2). The regulation identifies high emitting vehicles and forces their owners to repair
the vehicles or face penalties such as license revocation. The regulation may also induce pre-test
substitution effects. Emissions testing raises the price of owning a highly polluting car. This may
encourage out migration of cars to counties that do not test emissions, increases scrappage of highly
polluting cars, and increased pre-test maintenance (see Kahn 1993).
A county that starts an emissions testing program enjoys a change in total emissions that is
expressed in equation (4).
A total emissions - J ^ (N*f(E) - NfiEfi (4)
Equation (4) says that implementing an emissions testing program changes the composition of the
fleet (its age structure) and changes the distribution of emissions within the fleet (reducing the percentage
of high emittors). I concentrate on the regulation's "total impact" without attempting to decompose it into
these separate effects. If the regulation is effective because it encourages out migration of cars of highly
pollutin cars to counties that do not test emissions, then the regulation may represent a "zero sum" game.
In this case, my study's estimates do not represent the program's impact if all counties start such
programs.
IV. The Empirical Model to Estimate Regulation's Impact
To evaluate any environmental regulation's impact on a certain geographic area's air quality
requires a proxy ing measure for the intensity of regulation within that geographical unit. Vehicle
emissions testing programs are an attractive regulatory program to study because I can easily quantify
their presence. In a given year, a county either does or does not have such a program. For a given state,
the emissions testing programs are standardized across counties.
To quantify the aggregate impact of the regulation, I model per-capita pollution levels as a
function of weather, regulation and a time trend. I include a time trend to control for changes in the fleet
composition over time. Over time, cars built before 1979 represent a smaller share of the total fleet. This
is important because vehicle emissions have been falling sharply with respect to model year. Table Two
provides evidence of this. Note the sharp decrease in hydrocarbon and carbon monoxide emissions for
1979 model year cars versus 1989 model year cars (see White (1982) and Kahn (1993)).
The empirical model is presented in equations (5-7);
l o g ( t y Z p = <t>y. + PJT, + yDit + * Trend, + eiJt (5)
Dit =lift> 1986 A Ym > S (6)
A l o g ( i y z p = * + PA Xit + YA DU + A eiJt O)
Yijt = site j in county i's pollution level at time t
Zit = county i's employment at time t
5
Xit = county i's time t weather variables
Dit = county i's time t regulatory dummmy
psij = site j ' s fixed effect
eijt = site j in county i's time t error term, iid
S = Clean Air Act Standard
Equation (5) models county per-capita pollution as a function weather, a time trend and
regulation. Note that equation (5) includes the site specific fixed effect psij. This can represent both
geographical fixed effects of county j and spatial sample selection. Since air quality varies within
counties, monitoring stations near highways may have higher readings than monitors in the suburbs even
though aggregate county car activity is the same for both sites. I model pollution in per-capita terms
because this takes care of the problem that the different counties have different numbers of cars tested.
Multicollinearity between employment and the Chicago fixed effect. I am implicitly assuming that a one
percent increase in employment increases pollution by one percent.
Regulation enters equation (5) as a dummy variable. A negative coefficient for Dit would indicate
that counties that have emissions testing programs suffer less pollution for a given level of economic
activity. By running state by state regressions, this framework can be used to rank the effectiveness of
different states' programs. The goal of estimating equation (5) is study whether controlling for climate
variables, pollution per-capita fell when the regulation was imposed. Equation (6) states the presence of
an emissions testing program in a given county at a time t is a function of air quality in that county in
1982. This decision rule was written into the Clean Air Act Amendments of 1977. It required states to
implement emissions testing programs in areas not in attainment of the ozone and carbon monoxide
standards by 1982. In a single cross-sectional regression, one cannot estimate county fixed effects. In
this case, the phi term in equation (5) will be part of the disturbance term. Since regulation is not
randomly assigned but is imposed on the counties with the highest pollution, the alpha term would be
positively correlated with the regulation variable. This would bias the estimate downward. Least squares
yields a lower bound on regulation's impact.
Below, I present estimates of equations (5) and (7) using least squares. Least squares yields
consistent estimates of the parameters in equation (1) because of state implementation delays. Illinois did
not begin its emissions testing program until 1986 but determined in 1982 which counties had to
implement the program, Dit equals zero for all counties before 1986. In 1986 and later years, the
regulatory variable is not correlated with the disturbance term in equation (1). The Illinois testing
program was delayed for years because of the inconvience it imposes on each car owner. Such regulation
would be politically unpopular because each voter believes that his emissions are inconsequential for
aggregate Chicago air quality. Each voter would like to "free ride" on others pollution reduction efforts.
The federal government had to threaten to remove Federal Highway Funds if states did not comply. Since
the emissions testing program did not begin in 1982, Aeijt is not correlated with the regulatory variable.
If the most polluted counties had started emissions testing programs in 1982 then the regulatory variable
would have been correlated with the disturbance term and least squares would yield a lower bound on
regulation's impact. If the disturbance term follows an AR(1), then least squares yields biased estimates
of the program's true effect. If the disturbance term is serially uncorrelated or can be modeled as a
MA(1) or MA(2), then least squares yields consistent estimates.
Unlike MacAvoy (1987), I control for the fact that the "dirtiest" counties implement the program.
Most regulatory evaluation studies, such as Fuchs (1986), assume that state regulation is exogenously
determined. Fuchs, using a short state panel, estimates levels regression to study the impact of vehicle
emissions testing on motor accident mortality. Without exploring why regulation varies spatially, one
cannot identify if it is the regulation or the characteristics of the states that pass such legislation that led
to lower motor accident mortality.
V. Data Sources
The air quality data are from the Illinois state EPA. For each monitoring site within a county,
the Illinois EPA provides the yearly empirical distribution for each pollutant it monitors. I focus on
carbon monoxide and ozone. Three measures of a county's yearly air quality are used: the median, the
99th percentile and the second highest yearly reading. Since the median is a more robust statistic, it will
be less sensitive to extreme weather patterns. Therefore, regulation should have a greater impact on this
measure of air quality than on the outlier indicators. The second highest yearly reading is used because
the EPA bases its demands for further county regulation stringency on trends in these statistics. By
concentrating on outliers, the EPA indicates that it is more worried about short exposure to high air
pollutant levels rather than long exposure to lower dosages.
For Illinois, a given monitoring station may take 8000 readings a year. The yearly second highest
reading is an extreme order statistic. A county's probability of having a second highest pollution reading
exceed a given standard is a non-decreasing function of how many air quality samples are taken during
the year. If larger counties have more monitoring stations than smaller counties then even if true air
quality is the same, the larger counties are more likely to have regulation imposed.
The Illinois air quality data may not be representative of the state's pollution levels because not
every location is monitored. The EPA collects air quality data to update its information on which
locations may be non-attainment areas. Given patterns of where the EPA does and does not monitor air
quality, it would appear that the EPA's goal is to minimize the number of people exposed to high air
pollution levels. In 1988 in Illinois, only 19 of 102 counties had their ozone levels measured and only
9 counties had carbon monoxide readings taken. A county's probability of being monitored is an
increasing function of its population size. For a given county, air quality is not monitored at random
locations. Instead, the most polluted areas are most likely to be sampled. For Illinois, I have 389 county
year observations for ozone, 135 for carbon monoxide. This data set covers all Illinois ozone and carbon
monoxide monitoring sites between 1980 and 1989. The Illinois vehicle emissions testing program started
in 1986 in five counties.2
Equation (5) indicates that the dependent variable is pollution per-capita. I use total employment
data by county by year from the Bureau of Economic Analysis's county level REIS (Regional Economic
Information System) data set. Weather data is available at a monthly basis from 1895-1989 from the
National Climate Information Disc from the Department of Commerce. Jones (1992) shows the
importance of accounting for weather patterns in studying ozone trends.
VI. Empirical Results
Air Quality Trends
Air quality trends differ depending on what air quality statistic is used. Consider the case of Los
Angeles county, known to have the worst ozone problem in the nation. Between 1978 and 1991, the
statistic the EPA uses for studying ozone trends, the second highest yearly reading, fell 26%. A more
robust statistic is the percentage of readings that exceeded the ozone standard of 125 parts per billion.
This index has decreased by 61% over the same period.
Table Three presents time series trends for Illinois carbon monoxide and ozone levels between
1980 and 1990. The carbon monoxide trends suggest that sharp improvement occurred during the 1980s.
In addition to time dummies, Table Three also includes two other regressors. The first is a dummy
variable that equals one if for the five counties that eventually start an emissions testing program.
Surprisingly, these counties have lower median, tail, and outlier carbon monoxide levels than counties
that do not start programs. I interpret this as evidence of spatial sampling variation. Air quality is not
homogeneous within large counties such as Cook county. It is possible that there are more monitors in
2 These counties are DuPage, Cook, Madison, Lake and St. Clair.
9
the Chicago suburbs than in the Peoria, Illinois suburbs. The second dummy indicates for a given county
in a given year whether it test emissions. Surprisingly, this variable is not statistically significantly
negative in any of the regressions. Thus, the trend data suggest that testing programs have not played an
important role in improving Illinois air quality. Note that for median ozone, air pollution has actually
increased over time. An interesting finding is that counties that start emissions testing programs had
lower median ozone levels than counties that did not. Note that counties that eventually start emissions
programs have a higher 99th percentile ozone level and a higher second highest yearly reading than
counties that do not start programs. The results suggest that 1988 was an extremely high year for ozone.
High temperatures that year drove ozone upward (see Jones 1992).
By way of comparison, air quality data for California in the 1980s indicates similar patterns.
Carbon monoxide problems have been solved before the state began the emissions testing program in
1984. For example, in 1973 only 0.4% of all carbon monoxide readings exceeded the National Ambient
Air Quality Standard of nine parts per million. By 1979 this percentage had decreased to 0.1%. Similar
to Illinois, less progress has been made in reducing California ozone (smog). Between 1980 and 1989,
ozone levels outside of Los Angeles did not change.
Emissions Program's Impact on Ambient Air Quality
This section presents estimates of equation (5) and equation (7) using Illinois data. Table Four
presents three regression estimates of equation (5) for Illinois ozone. A separate fixed effect is estimated
for each monitoring site. Since there were 64 monitoring stations, I suppress these estimates. There was
significant variation in these estimates. Within Chicago, different monitoring stations had very different
intercepts.
The left column of Table Four presents results when the dependent variable is the log of a
10
county's per-capita median ozone. The middle column presents results for the log of county's per-capita
99th percentile of the yearly ozone distribution and the right column presents results for the log of a
county's per-capita second highest reading from the yearly ozone distribution. Following equation (1),
the independent variables include a fixed effect for each county in the data set, a time trend, weather
variables, and the regulation dummy. The key finding is that for these log-level regressions, the
emissions testing program does have a statistically significant impact on ambient ozone. For all three
indicators of air quality, the program reduced ozone by approximately 11%. This estimate is intuitively
plausible. Note that the weather variables estimates are quite intuitive for the tail and outlier regressions.
Increased July temperature and decreased rain both have a statistically significant impact on increasing
the 99th percentile of the ozone distribution and on the outlier measure. Interestingly, the weather
variables are not an important component for explaining median county ozone. The time trend variable
indicates that median per-capita ozone is increasing at 3 % a year while the 99th percentile and the outlier
are decreasing at 2% and 3%, respectively, per year. The very high R2 in the regressions are the result
of having estimated fixed effects.
Table Five presents estimates of equation (7), growth rate regressions, for ozone. First
differencing the data removes the county fixed effects and the monitoring site specific effects. The
differenced results suggest that the regulation has lowered ozone. The median per-capita growth rate fell
35%. The 99th percentile growth rate fell 14% and the ozone outlier growth rate fell 20%. Note that
similar to the ozone levels regressions, the weather variables are statistically significant with the expected
signs.
Table Six presents estimates of regulation's impact on Illinois carbon monoxide (CO) levels. The
Table has the same structure as Table Four. Unlike the ozone estimates, I find weaker evidence that the
emissions testing regulation has been effective. My estimates suggest that the program has reduced
median per-capita carbon monoxide by 11 % and outlier carbon monoxide by 7 %, and has increased the
11
99th percentile of the carbon monoxide distribution by 5 %. None of these estimates are significant at the
5% significance level. The Illinois EPA claims that the program has led to a 25% reduction in carbon
monoxide emissions. Note the negative and statistically significant time trend for all three regressions
reported in Table Six. Median CO levels are falling 7% a year. The 99th percentile of CO is falling 8%
a year and the outlier is falling 6% a year. These findings suggest that changes in fleet composition, i.e
the low emissions of newer cars, not the reduced emissions of existing cars is the key to reduced CO
levels in Illinois. Note the weather variables' impact. Increased rain lowers CO levels while July weather
has no impact. Table Seven presents the difference estimates for carbon monoxide. I find very similar
results as in the levels estimates in Table Six. The emissions test's presence does not lower carbon
monoxide levels. Increased rainfall still lowers CO. Note the constant. It represents the time trend. It is
negative and statistically significant. The constant's estimates in the three regressions are roughly the
same as the time trend estimates in Table Six.
To summarize, I have found evidence that the emissions program has reduced ozone pollution
but not had a signficant impact on carbon monoxide emissions. My estimates of the program's impact
are smaller than the EPA has claimed.
VII. Conclusion
Little research has investigated whether EPA regulation has been effective at improving county
air quality. This paper has studied the impact of the major regulation aimed at lowering existing mobile
source emissions. I found suggestive evidence that the Illinois vehicle emissions testing program has
lowered ozone but has not achieved the claimed reductions in carbon monoxide pollution. The time trend
estimates for carbon monoxide suggest that new car emissions reduction may be responsible for the
improvements in carbon monoxide during the 1980s. This is an important finding because the current
12
emissions testing regulation is costly in terms of time and administrative costs. If each new model year's
aggregate emissions continue to fall, the whole emissions testing regulation could be abolished in ten
years.
Thie paper's findings are relevant for evaluating the benefits of the Clean Air Act Amendments
of 1990 that mandate more stringent emissions testing (the IM240) at locations that are still not in
compliance with Clean Air Act air quality standards. A cost/benefit analysis of adopting this more
stringent regulation must include a study of forgoing the current idle test (for more on the IM240 see
McConnell and Harrington 1992 and the U.S General Accounting Office 1992).
This paper used cross-section and temporal variation in county regulatory status to quantify the
impact of emissions testing. I estimated the program's impact while controlling for the fact that regulation
is not randomly assigned. A researcher who ignores this information and assumes that the regulation was
exogenously imposed would underestimate the program's impact. Since air quality is an output measure,
future research could rank states by their program's effectiveness.
Other environmental regulatory efforts could be evaluated using regression techniques if the
decision rule for enforcement intensity was well understood. For example, it would be interesting to know
what has been the impact of Federal Air Grant Transfers on state and county air quality. Deily and Gray
(1991), Magat et. al. (1986) have studied what factors influence federal environmental regulators'
monitoring and enforcement decisions.
13
REFERENCES
California Environmental Protection Agency. Ozone Air Quality Trends 1981-1990. June 1992.
California I/M Review Committee. Evaluation of the California Smog Check Program andRecommendations for Program Improvements. Fourth Report to the Legislature. October 15,1992.
Deily, Mary and Wayne Gray. "Enforcement of Pollution Regulations in a Declining Industry."Journal of Environmental Economics and Management. (1991) pp. 260-274.
Florida Department of Highway Safety. "Annual Report 1992: Florida's Motor Vehicle InspectionProgram." 1992.
Fuchs, Victor and Irving Leveson. "Motor Accident Mortality and Compulsory Inspection of Vehicles."The Health Economy. Harvard University Press 1986.
Heckman, James. "Simultaneous Equations Model with Continuous and Discrete Endogenous Variablesand Structural Shifts." In Studies in Non-Linear Estimation. Edited by S.M. Goldfield an R.E.Quandt. Cambridge MA: Ballinger. 1976
Jones, K. H. "The Truth About Ozone and Urban Smog." Policy Analysis. 1992.
MacAvoy, Paul. "The Record of the Environmental Protection Agency in Controlling Industrial AirPollution." In Energy Markets and Regulation. Edited by R.L. Gordon et. al. Cambridge MA:Ballinger 1987, pp. 459-488.
MacAvoy, Paul. Industry Regulation and the Performance of the American Economy. New York. W.W.Norton. 1992.
Magat, Wesley and Alan Krupnick and Winston Harrington. Rules in the Making: AStatistical Analysis of Regulatory Agency Behavior. Resources for the Future. WashingtonD.C. 1986.
McConnell, Virginia and Winston Harrington. "Cost-Effectiveness of Enhanced Motor VehicleInspection and Maintenance Programs." Resources for the Future Discussion Paper QE92-18-REV, 1992.
MVMA Motor Vehicle. Facts and Figures' 92. Motor Vehicle Manufacturers Association of theUnited States Inc. Detroit 1992.
United States General Accounting Office. Unresolved Issues May Hamper Success of EPA's ProposedEmissions Program. GAO/RCED-92-288. September 1992.
14
Table One
Summary Statistics
Variable
log of (median carbonmonoxide/total countyemployment)
log of (tail carbonmonoxide/total countyemployment)
log of (outlier carbonmonoxide/total countyemployment)
log of (median ozone/totalcounty employment)
log of (tail ozone/total countyemployment)
log of (outlier ozone/ totalcounty employment)
July temperature
yearly rainfall
emissions testing dummy
observations
135
135
135
389
389
389
389
389
389
Mean
-6.18
-4.51
-3.82
-9.26
-7.94
-7.56
74.4
48.3
0.21
Std. Dev.
1.88
1.77
1.79
1.94
1.82
1.80
1.94
1.17
0.40
15
Table Two
Chicago Cars Emissions and Their Pass Rates
Model Year
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
pass rate
88
87
86
84
87
84
87
81
70
72
80
83
84
88
93
94
95
hydrocarbons
median
47.5
66
40
37
36
40
37
18
31
28
17
17
10
6
3
3
2
75th percent
145
244
210
150
147
164
158
113
112
90
58
57
50
42
27
23
20
carbon monoxide
median
1.35
1.42
1.53
1.83
2.11
2.04
1.69
0.73
0.64
0.70
0.63
0.60
0.51
0.52
0.45
0.40
0.38
75th percent
3.31
2.89
3.24
3.70
3.92
3.77
3.42
1.83
1.57
1.58
1.37
1.33
1.29
1.29
1.12
1.12
1.01
This table presents information on a data set of 50,000 cars whose emissions were tested during 1992 in Chicago. For eachmodel year between 1973-1989,1 present the mean percent of cars that passed the emissions test. I also present, by model year,the median and 75 percentile of the empirical emissions distribution for both hydrocarbons and carbon monoxide. Hydrocarbonsare measured in parts per million. Carbon monoxide is measured as a percent of the exhaust gas.
16
Table Three
Illinois Air Quality Trends in the 1980s
constant
1981 year dummy
1982 year dummy
1983 year dummy
1984 year dummy
1985 year dummy
1986 year dummy
1987 year dummy
1988 year dummy
1989 year dummy
1990 year dummy
emission program
emission program impact
obs/ R squared
Median CO
1.61(12.8)
-0.19(-1.28)
-0.36(-2.29)
-0.28(-1.76)
-0.28(-1.65)
-0.44(-2.80)
-0.44(-2.75)
-0.34(-1.94)
-0.45(-2.84)
-0.53(-3.40)
-0.60(-3.87)
-0.68(-3.88)
0.35(2.04)
158 0.28
Outlier CO
13.5(12.8)
-0.25(-0.20)
-1.72(-1.31)
-0.72(-0.54)
0.72(0.52)
-2.72(-2.03)
-3.03(-2.32)
-1.69(-1.16)
-3.06(-2.32)
-3.56(-2.73)
-3.93(-3.00)
-2.25(-1.52)
-0.03(-0.02)
158 0.26
Median Ozone
0.020(24.8)
0.001(1.05)
0.002(1.73)
0.004(3.52)
0.003(2.80)
0.006(5.73)
0.002(1.29)
0.003(2.30)
0.007(5.86)
0.006(5.25)
0.004(3.81)
-0.005(-7.68)
0.002(1.95)
433 0.30
Outlier Ozone
0.10(25.0)
-0.005(-0.93)
-0.018(-3.22)
0.019(3.41)
0.002(0.44)
-0.003(-0.51)
-0.013(-2.22)
0.012(1.94)
0.023(3.83)
-0.009(-1.50)
-0.021(-3.55)
0.017(5.51)
0.003(0.55)
433 0.32
Note: t-statistics in (). Ozone is measured in parts per billion. Carbon monoxide is measured in parts per million.The independentvariables include year dummies with 1980 being the omitted category, emission program is a dummy variable that equals one ifthe county ever starts an emissions testing program. Emission program impact is a dummy variable that equals one if a county hasstarted a program.
17
Table Four
Impact of Emissions Testing Program on Illinois Ozone Levels
Dependent Variable: the log of the per-capita ozone measure
emissions test
July temperature
yearly rainfall
time trend
observations
R squared
Median
-0.13(-2.72)
0.002(0.29)
0.007(1.86)
0.03(4.15)
378
0.99
Tail
-0.10(-2.46)
0.02(3.92)
-0.01(-3.33)
-0.02(-3.72)
378
0.99
Outlier
-0.11(-1.89)
0.03(3.1)
-0.01(-2.13)
-0.03(-3.22)
378
0.98
Note: t-statistics are in parentheses. Median represents the median of a county's yearly empirical ozone distribution. Tail represents the99th percentile of this distribution. Outlier represents the second highest yearly reading. The independent variables include a timetrend, the regulation dummy, July temperature and yearly rainfall. A separate fixed effect was estimated for each of the 62 monitoringsites in the data set. In this table, I do not present the fixed effect estimates. I estimate this regression using least squares on 1980-1989data.
18
Table Five
Impact of Emissions Testing Program on Illinois Ozone Growth
Dependent Variable: the growth rate of the per-capita ozone measure
emissions test
July temperature
yearly rainfall
constant
observations
R squared
Median
-0.35(-4.16)
0.46( 0.86)
0.12(1.03)
0.05(2.45)
318
0.06
Tail
-0.14(-1.86)
1.52(3.30)
-0.27(-2.56)
-0.03(-1.47)
318
0.07
Outlier
-0.20(-2.09)
1.73(2.84)
-0.28(-2.00)
-0.02(-1.06)
318
0.06
Note: t-statistics are in parentheses. Median represents the median of a county's yearly empirical ozone distribution. Tail represents the99th percentile of this distribution. Outlier represents the second highest yearly reading. The independent variables include a time trendthe regulation dummy, July temperature, yearly rainfall, using least squares on Illinois county data from 1980-1989. Fixed effectsmodel where I do not present the county specific intercepts.
19
Table Six
Impact of Emissions Testing Program on Illinois Carbon Monoxide Levels
Dependent Variable: the log of the per-capita carbon monoxide measure
emissions test
July temperature
yearly rainfall
time trend
observations
R squared
Median
-0.11(-1.31)
-0.005(-0.40)
-0.02(-2.65)
-0.07(-5.16)
133
0.99
Tail
0.05(1.26)
0.001(0.09)
-0.006(-1.67)
-0.08(-11.4)
133
0.99
Outlier
-0.07(-1.54)
-0.01(-0.97)
-0.01(-2.05)
-0.06(-5.18)
133
0.99
Note: t-statistics are in parentheses. Median represents the median of a county's yearly empirical carbon monoxide distribution. Tailrepresents the 99th percentile of this distribution. Outlier represents the second highest yearly reading. A separate fixed effect wasestimated for each of the 26 monitoring sites in the data set. In this table, I do not present the fixed effect estimates. I estimate thisregression using least squares on 1980-1989 data.
20
Table Seven
Impact of Emissions Testing Program on Illinois Carbon Monoxide Growth
Dependent Variable: the growth rate of the per-capita carbon monoxide measure
emissions test
July temperature
yearly rainfall
constant
observations
R squared
Median
0.07( 0.67)
-0.14(-0.18)
-0.32(-1.86)
-0.09(-3.23)
105
0.04
Tail
0.02(0.31)
0.03(0.07)
-0.11(-1.16)
-0.06(-4.05)
106
0.02
Outlier
-0.05(-0.49)
-1.09(-1.38)
-0.30(-1.65)
-0.05(-1.73)
106
0.04
Note: t-statistics are in parentheses. Median represents the median of a county's yearly empirical carbon monoxide distribution. Tailrepresents the 99th percentile of this distribution. Outlier represents the second highest yearly reading. The independent variablesinclude a time trend the regulation dummy, July temperature, yearly rainfall, using least squares on Illinois county data from 1980-1989. A fixed effects model is estimated. I do not present the monitoring site specific intercepts.
21
1993-94 Discussion Paper SeriesDepartment of Economics
Columbia University420 W. 118 St., Room 1022
New York, N.Y., 10027Librarian: Angie Ng
The following papers are published in the 1993-94 Columbia University Discussion Paperseries which runs from November 1 to October 31. Domestic orders for discussion papers areavailable for purchase at $5.00 (U.S.) each and $140.00 (U.S.) for the series. Foreign orderscost $8.00 (U.S.) for individual paper and $185.00 for the series. To order discussion papers,please send your check or money order payable to Department of Economics, ColumbiaUniversity to the above address. Please be sure to include the series number for the paper whenyou place an order.
671. Investment in U.S. Education and TrainingJacob Mincer ( Nov. 1993)
672. Freer Trade and the Wages of the Unskilled: Is Marx StrikingAgain?Jagdish Bhagwati and Vivek Dehejia
673. Employer Size and Labor TurnoverTodd Idson
674. Less Crime May Be WorseBrendan O'Flaherty
675. Team Production Effects on EarningsTodd Idson
676. Language, Employment, and Earnings in the United States:Spanish-English Differentials from 1970 to 1990David Bloom and Gilles Grenier
677. The Impact of Performance Incentives on Providing Job Trainingto the Poor: The Job Training to the Poor: The Job Training PartnershipAct (JTPA)Michael Cragg
678. The Demands to Reduce Domestic Diversity among Trading NationsJagdish Bhagwati
679. Mass Layoffs and UnemploymentAndrew Caplin and John Leahy
680. The Economics of AdjustmentAndrew Caplin and John Leahy
681. Miracle on Sixth Avenue: Information Externalities and SearchAndrew Caplin and John Leahy
682. Arbitrage, Gains from Trade and Scoial Diversity: A Unified Perspective onResource AllocationGraciela Chichilnisky
683. Who should abate carbon emissions?Graciela Chichilnisky, Geoffrey Heal
684. Believing in Multiple EquilibriaGraciela Chichilnisky
685. Limited Arbitrage, Gains from Trade and Arrow's TheoremGraciela Chichilnisky
686. International Emission Permits: Equity and EfficiencyGraciela Chichilnisky, Geoffrey Heal and David Starrett
687. Do Vehicle Emissions Testing Program Improve Air Quality?Matthew Kahn
688. Sources of Real Exchange Rate Fluctuations: How Important Are Nominal Shocks?Richard Clarida and Jordi Gali
689. Modeling Soviet Agriculture for Assessing Command Economy PoliciesPadma Desai and Balbir Sihag
690. The Changing Labor Market Position of Canadian ImmigrantsDavid Bloom, Gilles Grenier and Morley Gunderson
691. Herd Behavior, the " Penguin Effect ", and the Suppression ofInformational Diffusion: An Analysis of Informational Externalitiesand Payoff InterdependencyJay Pil Choi
692. Shock Therapy and Russia: Was It Tried? Why Did It Fail? What Did ItDo? What Now?Padma Desai
693. Changes in the Structure of Family Income Inequality in the United Statesand Other Industrial Nations during the 1980sMcKinley L. Blackburn and David E. Bloom
694. Necessarily Welfare-enhancing Customs Unions with IndustrializationConstraints: a Proof of the Cooper-Massell-Johnson-Bhagwati ConjecturePravin Krishna and Jagdish Bhagwati
695. The Effect of Household Characteristics on Household-Specific InflationRates: An Application to Trends in Child Poverty and Educational RealWage DifferentialsTodd Idson and Cynthia Miller
696. Threats to the World Trading System: Income Distribution and theSelfish Hegemon
697. Intraindustry Trade: Issues and TheoryJagdish Bhagwati and Donald R. Davis